We've accidentally built a backend framework for Large Language Models (LLMs) that simplifies API orchestration. This framework allows developers to build modular monoliths, microservices, or a mix of both, and is inspired by existing frameworks like Vercel's Lovable, Bolt, and Backstage. The framework leverages LLMs to generate code for proxying between the Supergraph and non-GraphQL services, reducing the cost of creating a unified API graph. Cosmo Plugins are the core of this framework, allowing developers to define schemas, compose modules, and generate code using LLMs like Cursor. This approach enables building truly modular monoliths with isolated plugins, making it easier to deploy and manage APIs. The framework is designed to be flexible, allowing developers to choose their own deployment strategy and handle dependencies between plugins in a declarative way.